Taming workflow chaos in generative design data

The mission: Fuel design exploration with AI-curated visual breadth

Our client was developing a cutting-edge Generative AI model designed to produce design assets across a broad spectrum of art styles. To train their model effectively, they needed a vast and highly diverse corpus encompassing everything from delicate watercolor textures to abstract and complex styles. A team was needed to source, curate, and prepare over 200,000 vector images spanning multiple artistic categories within an exceptionally tight timeframe.

The challenge: Orchestrating diverse image data workflows under pressure

The client faced multiple critical challenges spanned across their data transformation journey.

Key obstacles:

• High volume of data required under tight timelines
• Wide-ranging artistic styles needed, further demanding nuanced labeling and segmentation
• Strict and abstract quality criteria complicated image selection and precise categorization
• Laborious sketching for ground truth sourcing
• Workflow complexity to streamline tasks from sourcing vectors and sketching art styles to multi-tag labeling and file exports

The goal:

Bridge siloed workflows to scale varied image data enablement with speed and precision.

The solution: Integrated workstream of process, experts and tools

We implemented a tightly integrated approach combining agile operational data processes, skilled creative experts, and customized tools to streamline every stage of our data delivery.

Our approach:

1. Designed a high-pressure workstream made to support visibility, collaboration, and asynchronous execution across multi-phase data tasks
2. Onboarded and trained a dedicated team of over 100 graphic designers that enabled thousands of images to be sourced, sketched, and labeled daily over several months
3. Implemented rapid iteration and feedback cycles for early anomalies detection and daily quality monitoring
4. Customized our proprietary platform to facilitate task assignments, progress tracking, data cleansing and validation, and data export tailored to client’s model

The results: AI-ready creative data at scale

• Data accuracy and consistency improved to above 96%
• Increased throughput volume by 20X compared to client’s previous efforts, from 10,000 to 200,000 images quarterly
• Accelerated time-to-market with faster model iterations and deployment
• Cost reduction with efficient resource allocation and process automation

This partnership transformed a complex, time-sensitive data challenge into a scalable, quality-driven pipeline that empowered the client’s Generative AI to create stunning design assets faster and more reliably than ever before.